Blind Estimation of Spectral Standard Deviation from Room Impulse Response for Reverberation Level Recognition Based on Linear Prediction

Abstract

Reverberation is an important factor affecting speech quality and intelligibility, Reverberation Time (RT) and Direct-to-Reverberant Ratio (DRR) are the primary parameters for reverberation strength judgement, spectral standard deviation (SSD) from room impulse response (RIR) and DRR exist as monotonic relationships to some extent which means that SSD can also be an indicator of reverberation characteristics. We propose a blind estimation of spectral standard deviation (BESSD) that is obtained directly from reverberant speech signals. Experiments prove BESSD can be used as an index for male reverberation level recognition.